Neural Reward Machines

Elena Umili, Francesco Argenziano, Roberto Capobianco

Non-markovian Reinforcement Learning (RL) tasks are very hard to solve, because agents must consider the entire history of state-action pairs to act rationally in the environment.…